The identifier wave

DOI, ISBN, and URL coverage by the age of the cited work

Every citation record carries the identifier of the medium that produced its source — an ISBN for a paperback, a DOI for a journal article, a URL for a web page. Plotted against the age of the cited work, those identifiers stop being cataloging trivia and become a history of how knowledge was packaged: each one rises, crests, and recedes as the next medium takes over the work of pointing at sources.

Figure 4. Share of citation records carrying a DOI, an ISBN, or a URL, by the decade the cited work was published: ISBN coverage crests at 28.5% for works of the 1990s, DOI at 11.8% for works of the 2000s, and URL reaches 93.2% for the newest works after a floor of 37.2% in the 1830s. Denominator: the 21 936 052 citation records of the enwiki-20260601 citation spine with a parseable four-digit cited year — rates are conditioned on parseable years, and 7 459 305 further rows carry a missing or unparseable one. The plotted window 1400–2020 holds 21 909 707 of those records (99.9%); 45 decades before 1400 (26 329 records in all) and six post-2020 buckets (at most five records each) are too thin to draw, and the strip beneath the lines counts records per decade on a log scale.

Source: K2-citeyears receipt (coverage_decade) over the citation spine full-v2, via the dataset page.

Three identifiers, three waves

ISBN is the sharpest wave. It stays in the low single digits for everything published before the 1950s — at most 7.2%, typically 1–4%, small but never zero — then climbs to 28.5% for works of the 1990s, the era when every paperback carried a barcoded ISBN and every bibliography copied it, and falls back to 2.0% for the newest works. DOI rides a lower, later swell: barely visible until the 1990s, cresting at 11.8% for works of the 2000s, down to 4.2% now. The URL line is the tide that lifted everything else: it never drops below 37% anywhere in the drawn window — the floor, 37.16%, sits in the 1830s — reaches 93.2% for the newest works, and its climb from the 1980s on is the web swallowing the reference list.

The recession of ISBN and DOI in the newest buckets is a mix shift, not identifier abandonment: citations to works of the 2020s point overwhelmingly at web pages and preprints, which carry neither. And the URL line's high plateau over old works — 54.4% to 90.9% for everything before 1700 — measures the citing, not the cited: citations to fifteenth-century incunabula are made today, by web pages, and carry the web page's address.

How to read it

Each decade bucket collects cited works published in years D through D+9. Rates are conditioned on a parseable four-digit cited year; the 7 459 305 spine rows without one are excluded from every number on this page. Within the drawn window the thinnest decade, the 1420s, rests on 254 records — the early decades wobble on hundreds to a few thousand records each, which the strip beneath the lines makes visible by counting them on a log scale. Hover or focus any decade column for its exact record count and all three rates.

The full 115-decade payload — including the gated pre-1400 decades, the post-2020 buckets, and the unparseable-year bucket — lives in the figure's data payload, described on the dataset page.